一种基于证据多视角的模糊C-means聚类算法  

A Fuzzy C-means Clustering Algorithm Based on Multi-view of Evidence

作　　者：马宗方[1] 李雷华 田鸿朋 MA Zongfang;LI Leihua;TIAN Hongpeng(College of Information and Control Engineering,Xi’an University of Architecture and Technology,Xi’an 710055,China;College of Electrical Engineering,Zhengzhou University,Zhengzhou 450007,China)

机构地区：[1]西安建筑科技大学信息与控制工程学院,陕西西安710055 [2]郑州大学电气工程学院,河南郑州450007

出　　处：《控制工程》2024年第8期1345-1354,共10页Control Engineering of China

基　　金：陕西省重点研发计划项目(2020GY-186,2020SF-367);西安建筑科技大学科技基金项目(ZR21034)。

摘　　要：针对传统多视角聚类算法难以准确识别噪声和有效划分类间重叠区域样本的问题,提出一种基于证据多视角的模糊C均值(evidential multi-view fuzzy C-means,EMVFCM)聚类算法。首先,在证据推理框架下,研究一种改进的模糊C-means多视角聚类算法,通过优化改进的目标函数获得待测样本属于单类和噪声的信任值,从而识别出噪声数据。然后,由于重叠区域的样本不能被准确地划分类别,所以将其划分到相对应的复合类,这不仅能够表征数据样本类别的不精确性,还能降低错误分类的风险。最后,通过人工数据集和UCI数据集验证本文算法的性能并与相关算法对比。实验结果表明,本文算法较传统多视角聚类算法能更有效地处理数据中的噪声和重叠样本难以准确划分的问题。Aiming at the problem that traditional multi-view clustering algorithm is difficult to accurately identify the noise and effectively classify the samples of overlapping area between classes,an evidential multi-view fuzzy C-means(EMVFCM)clustering method is proposed.Firstly,an improved fuzzy C-means multi-view clustering algorithm is studied under the framework of evidential reasoning.By optimizing the improved objective function,the confidence value of the tested samples belonging to single clusters and noise can be obtained,and the noisy data can be accurately identified according to the clustering results.Then,because the samples in overlapping areas can’t be classified accurately,they are divided into corresponding meta-cluster,which can not only represent the inaccuracy of data sample categories,but also reduce the risk of misclassification.Finally,the performance of the proposed algorithm is verified by artificial data sets and UCI data sets and compared with relevant algorithms.Experimental results show that the proposed algorithm can deal with the problems of noise and overlapping samples in data more effectively than traditional multi-view clustering algorithm.

关 键 词：多视角聚类 重叠区域 证据推理 复合类 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]